Oracle Manipulation Attacks

How Protocols Read Prices

An oracle manipulation attack starts from a design choice. A lending market needs a price to decide how much you may borrow against collateral and when to liquidate you, and a perpetual futures venue needs one to settle profit and loss. Whatever supplies that price is the oracle. Some protocols read a feed aggregated across many markets. Others read the spot price, the current trading price, straight from one exchange or one liquidity pool, because it is cheap and always there.

That second choice is the weak one. In 2019 the researcher samczsun showed that a protocol reading an on-chain price without validating it lets an attacker "manipulate the apparent price at a specific instant with minimal to no loss in slippage fees." Chainalysis estimates that decentralised finance protocols lost 403.2 million dollars to 41 oracle manipulation attacks in 2022 alone.

How an Oracle Manipulation Attack Works

The pattern has three steps: take a position whose value depends on the price, push the price on the thin market the oracle reads, then cash out before arbitrage pulls it back. Thin liquidity is what makes it pay. In a market with little depth, a modest purchase moves the price a long way.

Flash Loan Funded Manipulation

The attacker does not even need the money. A flash loan is borrowed and repaid inside a single transaction, and Aave's documentation explains that if it is not repaid, the whole transaction is reverted. That lets someone with little capital command millions for a few seconds. In October 2020 Harvest Finance said an attacker had "manipulated prices on one money lego (curve y pool) to drain another", repeatedly, using flash loans to take 33.8 million dollars.

Lending Markets and Perpetual Futures

In a lending market, inflated collateral lets the attacker borrow real assets and leave, and the protocol keeps the bad debt. On a derivatives venue, a moved reference price marks the attacker's positions as profitable or pushes other traders into liquidation.

Mango Markets is the best documented case. The CFTC alleged that in October 2022 Avraham Eisenberg built large leveraged positions in a swap tied to the MNGO token, bought MNGO rapidly on the three exchanges feeding the oracle, then used the inflated value as collateral to withdraw over 110 million dollars. A jury convicted him in 2024, but the trial judge overturned those convictions in May 2025, citing venue and insufficient evidence of misrepresentation. Treat it as a documented mechanism, not a settled verdict.

Leverage means the push can be tiny. In October 2026 a trader on Papertrade, which priced trades at the midpoint of another exchange's best quotes and offered up to 1,000x leverage, moved that quote about 0.12 percent and booked 1.28 million dollars of profit.

Defences Against Oracle Manipulation

• Time-weighted average prices. A TWAP averages the price over a window. Uniswap's documentation says the cost of manipulating one rises roughly in line with both pool liquidity and window length. The price is that a TWAP lags genuine moves.
• Several independent sources. A median across many venues means one pushed market cannot drag the result far. Chainlink's feed selection guidance tells developers to check that an asset has enough liquidity to resist manipulation.
• Circuit breakers and bounds. The same guidance lists "value bounds, caps, circuit breakers, freshness checks" as safeguards. A deviation limit rejects a price that jumps too far between reads, and caps on position size shrink the payoff.

The practical rule: before depositing, ask where a protocol gets the price of each collateral asset and how deep that market is. A thinly traded token accepted as collateral is an open invitation, and our exploit tracker records several recent losses that were price manipulation on a thin market rather than contract bugs.